How the numbers are worked out
A photon budget. Every pixel collects light from the target and light from the sky, plus noise the camera adds on its own. Integration time is the point where the target's signal stands far enough above the combined noise.
The model
For a total integration T made of subs of length t, the per-pixel signal-to-noise ratio is
SNR = S·T / √( S·T + B·T + (T/t)·(RN² + D·t) )
where S is the target's photo-electrons per second per pixel, B the sky's, RN the read noise per sub and D the dark current. Solving for T at a chosen SNR:
T = SNR² · (S + B + RN²/t + D) / S²
That one line is the whole calculator. Everything else is working out S and B.
Target signal
Each object's mean surface brightness comes from its catalogue magnitude and size: SB = mag + 2.5·log10(area in arcsec²). That mean is often misleading: a nebula's catalogue magnitude may belong to the embedded star cluster, a galaxy's size may include a faint outer disc nobody images, a planetary's tiny bright ring may sit inside a huge faint halo. For those objects the effective figure is pinned by hand, calibrated against what smart-telescope owners report for the structure they are actually trying to capture, and the target page shows the value used. Open clusters get a nominal value: their stars are point sources and the goal is colour, not depth.
Photons per pixel then follow from the aperture area, the pixel scale squared, a mag-0 flux of 1000 photons/s/cm²/Å, 1000 Å of bandwidth per colour channel and a combined efficiency of 0.5 for sensor QE, optics and filters.
Sky background
| Bortle | Description | Sky brightness (mag/arcsec²) |
|---|---|---|
| 1 | Excellent dark site | 22.0 |
| 2 | Typical dark site | 21.8 |
| 3 | Rural | 21.5 |
| 4 | Rural / suburban transition | 21.0 |
| 5 | Suburban | 20.5 |
| 6 | Bright suburban | 19.8 |
| 7 | Suburban / urban transition | 19.1 |
| 8 | City | 18.5 |
| 9 | Inner city | 17.8 |
Zenith, no Moon. A first-quarter Moon adds roughly one Bortle class; a full Moon two or three. Targets low in the sky suffer both extinction and brighter skyglow, so the rural figures assume the object is reasonably high.
Filters
A dual-band filter is modelled as passing 15% of the skyglow, 70% of an emission-line object's light and 15% of a broadband object's. That is why it is a large win on nebulae and a large loss on galaxies and clusters.
Quality levels
| Level | Per-pixel SNR | What it looks like |
|---|---|---|
| Quick look | 3 | Target obvious, still noisy |
| Solid | 6 | Clean enough to share |
| Showpiece | 12 | Smooth faint detail, print-worthy |
SNR is measured per pixel at the scope's native scale for the effective surface brightness above. Downsampling, binning and noise reduction all make an image look better than its per-pixel SNR suggests, which is why "quick look" is usable at all. Two adjustments: objects smaller than 15′ are viewed at full resolution rather than downsampled, so their SNR target rises by up to 1.5×; and every level has a minimum sub count (10, 30 and 90) because stacking software needs frames to reject satellites, wind and bad tracking, whatever the maths says.
Sky brightness from your location
"Use my location" estimates your zenith sky brightness from satellite night-lights data. The source is the VIIRS Day/Night Band annual composite published by NASA (public domain, via the GIBS tile service), resampled to about 2 km. Every lit cell is treated as a light source and its glow spread over the surrounding 250 km with a smooth version of Walker's law (glow falling as distance to the power 2.5, flattened within 3 km). The sum is added to a natural sky of 22.0 mag/arcsec² and the two free parameters are fitted to published sky-meter readings at 25 sites from central London to Death Valley. Expect the estimate to land within one Bortle class of a real sky-meter reading; terrain, coastlines, LED conversions since the composite and the current state of your local streetlights all move it. It is a planning number, and you can always pick a class by hand instead.
Your position never leaves your browser as such: the page fetches one pre-computed 1° × 1° tile (2,500 bytes) and reads the cell you are in. The server sees only which degree square was requested and keeps no access log for this site.
What the model ignores
- The Moon, haze, altitude and extinction. Plan on the darker side of your Bortle class only on a good night.
- Frames the app rejects. Wind, satellites and poor tracking cost real subs; add 10 to 20% in practice.
- Field rotation in Alt-Az mode, which crops the stacked edges over a long session.
- Colour balance. The figures are for a single Bayer channel; a red emission target is mostly red-channel data.
- Your taste. "Solid" is a judgement, not a law.
Read noise figures are typical high-gain values for each sensor, not measurements of individual units. Catalogue magnitudes and sizes are the commonly published values and are approximate; corrections are welcome via the address on the about page.